1. Field of the Invention
The present invention relates to image processing, and more particularly to an advantageous object extraction device, object extraction method, and recording media storing an object extraction program used in the extraction of an object image.
2. Description of the Related Art
The object extraction methods described below are known conventional methods for obtaining an object extraction image by extraction only an object from a plurality of photographed images in which the same object is photographed along with a background.
First, when the photograph image (referred to hereinafter as the “central image”) of the object that will serve as the object extraction image and the photograph images (referred to hereinafter as the “reference images”) of the object having parallax with respect to the central image are output, the central image is partitioned into bands in the horizontal direction, and the image partitioned into bands is further partitioned into segments that are enclosed by the edge points of the object. Next, the location of the corresponding points that correspond to the end point of each of the segments are searched for on an epipolar line, and as a result the depth is estimated based on the obtained parallax. Next, based on this estimated depth information, each of the segments of the central image is partitioned into foreground segments and background segments. Next, the set of partitioned segments in the foreground is obtained as the object extraction image. However, in the case that one part of this image is lost when this object retraction image is formed, if there is an effective segment in a range sufficiently close when viewed vertically, the lost segment is interpolated by pixels in the horizontal and vertical directions.
Moreover, in this object extraction method, the calculation parameters such as the number of pixel spans during each of the calculations and the brightness difference used for edge identification is set in advance.
Furthermore, in the case that a background part is extracted as the object, the set of segments grouped in the background segments may serve as the object extraction image.
FIG. 15 is a block diagram showing an example of the structure of a conventional object extraction device. In this figure, reference numeral 1 is the central camera that photographs the central image P1, reference numerals 2-1 and 2-2 are reference cameras that photograph reference images P2-1 and 2-2, and reference numeral 3 is an object extraction calculating unit that finds the object extraction image P3 by carrying out an extraction calculation of the object image on the input central image P1, and reference images P 2-1 and 2-2 according to the above-described object extraction method by using a calculation parameter set in advance. In addition, this object extraction calculating unit 3 is a conventional object extraction device.
Moreover, the conventional object extraction device is realized by large-scale computers, etc., such as a work station into which an object extraction program that uses the above-described convention extraction method, for example, is input.
However, in the above-described object extraction device, in the case that a part of the object extraction image is lost, in the interpolation processing using pixels in the horizontal and vertical directions, there is the problem that the locations that should be background image are mistaken for a foreground segment and made a part of the object extraction image, or an object extraction image expanded beyond the actual object is formed, a location that should be a foreground object extraction image is mistakenly made a part of the background image, and an object extraction image having a blank part is formed.
In addition, when the object extraction calculation is carried out by changing to a more sophisticated calculation parameter in order to resolve the above-described problems, the amount of the calculating increases, and more time is consumed to find the object extraction image. As a result a large-sale computer is necessary for realizing the conventional object extraction device.